OpenAI has revealed its first custom-designed processor, called Jalapeño, built in partnership with Broadcom to power the servers that handle ChatGPT requests and other large language models. The chip represents a significant step toward reducing the company's dependence on Nvidia's graphics processing units (GPUs), which have been in short supply as demand for AI infrastructure has skyrocketed.

Jalapeño is an ASIC, which stands for Application-Specific Integrated Circuit. Unlike general-purpose chips, an ASIC is engineered for a single task: in this case, AI inference. Inference is the process where a trained AI model processes a user's request and generates a response, such as when you ask ChatGPT a question. This differs from AI training, where models consume vast amounts of data to learn patterns and improve their capabilities.

How Does Jalapeño Compare to Competitors?

According to Broadcom CEO Hock Tan, Jalapeño matches the performance of two industry-leading chips: Nvidia's Blackwell processors and Google's Tensor processing units (TPUs). This is a notable achievement, as Nvidia has long dominated the AI chip market. However, other major tech companies have also begun developing custom chips to reduce their reliance on Nvidia.

• Microsoft: Has launched custom-designed AI chips to power its servers for both training and inference workloads.
• Meta: Developed its own AI processors to support its large-scale AI infrastructure needs.
• Amazon: Created custom chips to optimize performance across its cloud services and AI applications.

Despite these efforts, Nvidia's chips still lead in overall performance across the industry. OpenAI's move signals that the company believes Jalapeño offers a compelling alternative for its specific use case: running inference on ChatGPT and similar models at scale.

What Makes Jalapeño Significant for OpenAI's Future?

OpenAI announced that Jalapeño is just the beginning of a broader strategy. The company describes it as the "first step in a multi-generation compute platform," meaning there will be follow-up versions designed to handle increasingly demanding AI workloads. OpenAI expects to deploy Jalapeño by the end of 2026.

Early testing shows that Jalapeño will deliver substantially better performance per watt compared to current state-of-the-art chips. This efficiency metric matters because running large AI models consumes enormous amounts of electricity. A chip that delivers the same performance while using less power translates directly into lower operating costs and reduced environmental impact.

The partnership with Broadcom, announced just nine months before this reveal, demonstrates how quickly the AI industry is moving. OpenAI recognized that relying solely on Nvidia for its computational needs created a bottleneck. Custom chips allow the company to optimize hardware specifically for its models and infrastructure, rather than adapting its systems to fit off-the-shelf components.

"While OpenAI is still measuring final performance, early testing shows that Jalapeño will deliver performance per watt substantially better than current state-of-the-art," OpenAI stated in its announcement.

OpenAI, Official Announcement

This shift reflects a broader trend in the AI industry. As companies like Microsoft, Meta, and Amazon have also developed custom chips, the market is moving away from a Nvidia-dominated landscape toward a more diversified ecosystem. For OpenAI, Jalapeño represents both a practical solution to supply constraints and a strategic move toward greater control over its computational infrastructure. The chip's arrival by year-end 2026 could reshape how efficiently the company runs ChatGPT and future large language models at scale.